Recently, the low-altitude economy relying on UAVs has been a hot topic due to the increasing demands under various scenarios, such as aerial delivery, aerial photography, etc. As such, UAV detection seems an inevitable technology behind the low-altitude economy, ensuring safety in the airspace and on the ground. Motivated by this, various detection technologies have been developed. Unlike radar and photoelectric detection, UAV radio frequency (RF) signals can identify targets over a large range. This is because UAV RF signals can be transmitted over long distances and received and detected. Artificial intelligence (AI) is commonly used for UAV signal detection due to its superior performance, such as ResNet signal detection. Although the accuracy of UAV signal detection based on residual neural network (ResNet) is very high, it sacrifices network and computational complexity to achieve the purpose. So, this paper proposes a lightweight framework, one-dimensional lightweight ResNet (1D-LW-ResNet), which balances performance and network complexity. The framework is designed to detect UAV RF signals in two stages. Initially, intuitive spectrogram features are extracted from UAV RF signals in the time-domain dataset using short-time Fourier transform (STFT) algorithms. Subsequently, the proposed 1D-LW-ResNet is considered for UAV signal detection tasks based on the spectrogram feature. It is worth noting that the proposed network achieves a 98.8% reduction in model parameters compared to ResNet, all while maintaining accuracy. Simulations verify the effectiveness of the proposed scheme.